SSBD Ontology: Bioimaging Metadata Framework

• SSBD Ontology is an OWL 2–based two-tier metadata framework that facilitates rapid data publication and richly annotated curation in bioimaging.
• It separates a core layer anchored in standard biomedical ontologies from an instance layer representing detailed dataset records.
• By aligning with REMBI and external vocabularies, the ontology supports advanced semantic queries and adheres to FAIR principles.

Searching arXiv for the SSBD Ontology paper and closely related naming variants. The SSBD Ontology is an OWL 2–based, two-tier metadata framework for the Systems Science of Biological Dynamics Database (SSBD) that is intended to make diverse, multidimensional bioimaging datasets interoperable and reusable. Its central design decision is a separation between a class-centric core layer, which references existing biomedical ontologies and supports reasoning and validation, and an instance layer, which represents actual imaging dataset information as RDF individuals explicitly linked to the core classes. This architecture is designed to reconcile two operational requirements within the same platform: rapid dataset publication with minimal metadata in SSBD:repository and value-added, richly annotated curation in SSBD:database. The ontology is aligned with REMBI, integrates external biomedical ontologies, and supports advanced semantic queries over projects, datasets, biosamples, imaging methods, instruments, dimensional metadata, and cloud-accessible image assets (Yamagata et al., 4 Aug 2025).

1. Scope, naming, and domain position

In its strict usage, SSBD Ontology denotes the ontology-driven framework for interoperable bioimaging metadata used by RIKEN’s Systems Science of Biological Dynamics Database. It is explicitly described as a two-tier metadata framework for heterogeneous metadata vocabularies across repositories and for the tension between rapid, minimally annotated data release and richly annotated, high-value datasets needed for discovery (Yamagata et al., 4 Aug 2025).

A recurrent source of confusion is terminological rather than conceptual. SSDOnt is a distinct ontology for Single-Subject Design studies, implemented in OWL 2 DL for representing phases, interventions, outcomes, participants, and results in single-case research (Berges et al., 2024). By contrast, SSbD in the smart-fab literature refers to Safe and Sustainable by Design, a governance framework involving trusted industrial data spaces, professional proxies, and compliance tokens, rather than a bioimaging metadata ontology (Liao et al., 8 Jun 2026). This suggests that references to “SSBD Ontology” should ordinarily be interpreted as the bioimaging ontology for SSBD, not as a synonym for either SSDOnt or SSbD.

The ontology is positioned as part of the foundingGIDE harmonization effort and as an infrastructure component for a Findable, Accessible, Interoperable, and Reusable data ecosystem within the bioimaging community (Yamagata et al., 4 Aug 2025).

2. Two-tier architecture and design rationale

The ontology is explicitly organized into a core layer and an instance layer. The core layer is a hierarchy of classes and properties defined in OWL 2 and anchored to OBO Foundry ontologies, designed for open-world reasoning, cross-ontology alignment, and SHACL validation. The instance layer encodes actual datasets and related entities as RDF individuals typed by the core classes and connected through object and data property assertions (Yamagata et al., 4 Aug 2025).

This architectural split is matched to the reorganized SSBD platform. SSBD:repository supports rapid publication at paper submission time with minimal metadata. Its schema centers on project, dataset, paper information, and person, and it uses simple, essential relationships and a small set of required fields. SSBD:database, by contrast, is a value-added tier in which datasets are enriched with detailed biological and imaging annotations using controlled vocabularies and external ontologies, enabling deeper reuse and semantic discovery (Yamagata et al., 4 Aug 2025).

The paper frames this as a deliberate balance between flexibility and rigor. Flexibility is provided by the instance layer, which can represent heterogeneous, dataset-specific facts, including different imaging modalities, variable instrument subcomponents, and multiple tags. Rigor is provided by the core layer through domain/range and cardinality patterns in OWL, together with SHACL shapes checking required fields for rapid-release records. A plausible implication is that the ontology is intended not merely as a vocabulary, but as a curation and validation scaffold for incremental metadata maturation.

The formal constraints illustrate this design. For example, Project is constrained by has_paper_information exactly 1 PaperInformation, has_dataset_output min 1 Dataset, and has_project_participant min 1 Person. Dataset is constrained by has_biosample_information min 1 BiosampleInformation, has_imaging_method_total_information min 1 ImagingMethodInformation, and has_dimension_information min 1 DimensionData (Yamagata et al., 4 Aug 2025).

3. Core classes, properties, and semantic coverage

The core schema covers bibliographic, biological, imaging, storage, and dimensional aspects of imaging datasets. The principal classes are Project, Dataset, PaperInformation, Person, BiosampleInformation, ImagingMethodInformation, ImagingInstruments, OME-NGFF-ZARR_Information, and DimensionData (Yamagata et al., 4 Aug 2025).

Project is the top-level container for related datasets and bibliographic context. Dataset is the imaging dataset entity and serves as the hub for links to biosample information, imaging method metadata, OME-Zarr locations, and dimensional data. PaperInformation stores title, DOI, PMID, URL, and MeSH topical descriptors. Person includes contributor and author records with identifiers such as ORCID, and the paper also lists J_Global_id and researchermap_id among the represented identifiers (Yamagata et al., 4 Aug 2025).

The biosample layer is carried by BiosampleInformation, which includes organism, strain, anatomical structure, cell type, cell line, genetic modifications, and reagents. Related classes such as GeneticModification and ReagentInformation capture fields such as gene_name, genotype, method, Ensembl_gene_ID, UniProt_ID, reagent_term, reagent_url, concentration, concentration_unit, fold_dilution, tag_uri, and oligo_primer (Yamagata et al., 4 Aug 2025).

The imaging layer is modeled through ImagingMethodInformation and ImagingInstruments. The ontology represents modality, detection method, illumination, contrast sources, resolution-enhancing method, visualization method, and imaged parameters. Instrument structure includes device_body, device_module, device_objective, detector, light_source, filter set, and device_dichroic. This enables representation of image-acquisition details such as laser wavelength, power, magnification, numerical aperture, and detector type, where available (Yamagata et al., 4 Aug 2025).

The storage and access layer is modeled by OME-NGFF-ZARR_Information, which links datasets to s3_endpoint_url, vizarr_url, thumbnail_url, and file_sets_url. Dimensional and sampling metadata are captured by DimensionData, including x_dimension_data, y_dimension_data, z_dimension_data, xyzct, and time_scale_data (Yamagata et al., 4 Aug 2025).

Representative object properties include has_dataset_output, has_paper_information, has_project_participant, has_biosample_information, has_imaging_method_total_information, has_instrument_information, has_ome_zarr_information, has_dimension_information, is_about_organism, is_about_strain, has_genetic_modification, has_reagent_information, has_detection_method, has_imaging_method_recorded_type, and has_tag. Notably, has_project_participant is declared as a sub-property of RO:0000057 from Relation Ontology, which situates participation semantics within a broader interoperable framework (Yamagata et al., 4 Aug 2025).

4. Alignment with external ontologies and metadata standards

A defining feature of the SSBD Ontology is its selective reuse of external ontologies and vocabularies. The core layer references OBO Foundry ontologies, including Gene Ontology (GO), Cell Ontology (CL), UBERON, Cell Line Ontology (CLO), NCBI Taxonomy (NCBITAXON), and MONDO. It also aligns with Relation Ontology (RO), FBbi, MeSH, and OME-NGFF / OME-Zarr (Yamagata et al., 4 Aug 2025).

These alignments are operational rather than merely declarative. is_about_organism can link a biosample to NCBITaxon:10090. has_imaging_method_recorded_type can link to FBbi:00000246 fluorescence microscopy. has_detection_method can link to FBbi:00000304 CMOS. has_tag can link to FBbi:00000477 VenusFP or FBbi:00000082 EGFP. is_about_strain can link to a JAX:000664 strain identifier. Paper records are enriched with MeSH Topical Descriptors via PubMed (Yamagata et al., 4 Aug 2025).

The ontology also states that SSBD metadata version 3 complies with REMBI, and the core layer is designed to cover REMBI fields. These include study-level identifiers and contacts, sample description, imaging instrument and modality, image acquisition parameters and dimensions, data access and formats, and publication links. The REMBI alignment is expressed through specific class mappings: Project and Person for contacts and identifiers; BiosampleInformation for organism, strain, anatomical structure, cell type, cell line, genetic modification, and reagents; ImagingMethodInformation and ImagingInstruments for modality and instrument descriptors; DimensionData for x/y/z and temporal scaling; OME-NGFF-ZARR_Information for access endpoints; and PaperInformation for DOI, PubMed, and MeSH (Yamagata et al., 4 Aug 2025).

This suggests that SSBD Ontology functions as a semantic superstructure over checklist-style metadata. Rather than replacing formats such as OME-Zarr, it adds the biological, bibliographic, and ontological relations needed for cross-dataset discovery and knowledge-graph integration.

5. Curation workflow, instance modeling, and query capabilities

The curation workflow is described in Section 2.2 of the paper. Domain curators populate spreadsheet templates, imaging experts review, and an ontologist translates the curated sheets into OWL/RDF, reusing OBO Foundry classes, establishing object properties, and generating instance-level assertions. The paper characterizes this as a curator–ontologist complementary workflow that bridges rapid data ingestion and later semantic enrichment (Yamagata et al., 4 Aug 2025).

The instance layer makes the ontology operational. A concrete example in the paper shows project-199 containing dataset-199-Fig2_BrainSliceRGB, linked to a biosample annotated with NCBITaxon:10090 Mus musculus and JAX:000664 C57BL/6J, to imaging metadata including FBbi:00000246 fluorescence microscopy and FBbi:00000304 CMOS detection, and to an OME-Zarr location with S3 endpoints and Vizarr URLs (Yamagata et al., 4 Aug 2025).

The ontology is deployed in Apache Jena Fuseki (4.7.0) with core and instance data, enabling competency queries. Three representative use cases are reported. First, a species-specific dataset search for Homo sapiens (NCBITaxon:9606) returned 12 human-derived datasets. Second, a cross-modality comparison for a shared strain retrieved datasets annotated with JAX:000664 (C57BL/6J) and listed distinct imaging modalities with Vizarr URLs for side-by-side comparison. Third, a federated cross-ontology query linked biosamples annotated with GO:0005739 (mitochondrion) to MONDO diseases defined via RO:0004020 “disease has basis in dysfunction of”, returning diseases such as MONDO:0004069 and MONDO:0044970 (Yamagata et al., 4 Aug 2025).

The paper further notes an operational workflow in which semantic discovery via SPARQL is followed by programmatic streaming of OME-Zarr images and BDZarr quantitative tables into Python/Napari for analysis, supporting cross-study phenotype screening and morphometrics. A plausible implication is that the ontology is intended to support both retrieval and downstream computational analysis, rather than acting solely as catalog metadata.

6. FAIR support, implementation status, and limitations

The ontology is presented as a FAIR-enabling infrastructure. Findability is supported by links between datasets, projects, and publications, a public SPARQL endpoint at https://knowledge.brc.riken.jp/bioresource/sparql, and indexing with VoID + DCAT metadata.ttl. Accessibility is supported through OME-Zarr S3 endpoints and Vizarr URLs. Interoperability is supported by dereferenceable URIs from OBO Foundry and FBbi, Relation Ontology alignment, OWL 2 axioms, SHACL shapes, and Linked Open Data federation with MONDO. Reusability is supported through REMBI-compliant coverage, detailed biological and imaging annotations, and constraints validated by HermiT (Yamagata et al., 4 Aug 2025).

The implementation details reported in the paper are specific. The ontology contains OWL 2 axioms and, as of 05/12/2025, 1,478 classes, 522 instances, 235 properties (116 object, 119 data). The reported tools are HermiT reasoner for consistency checking and Apache Jena Fuseki 4.7.0 for the SPARQL endpoint. Availability is stated through NCBO BioPortal under CC BY 4.0, a GitHub repository with queries and usage notes, a read-only SPARQL service online, and long-term preservation in Zenodo with a concept DOI (10.5281/zenodo.15700645). The project uses semantic versioning, and every tagged release is said to be validated by HermiT and archived in Zenodo for reproducibility (Yamagata et al., 4 Aug 2025).

The paper also states several limitations and future directions. It notes that legacy RDF (pre-2018) lacked hierarchical structure and property constraints, and that the new ontology addresses this, but that continued harmonization with other repositories (IDR, BIA) remains necessary, including differences such as ChEBI vs PubChem for chemicals. It further states that disease knowledge is not embedded directly, although federation with MONDO provides a practical workaround. Planned work includes LOV registration, broader dataset coverage, RDF-star for compact provenance, LLM/VLM-assisted annotations, fine-grained lineage metadata, and enhanced cross-database compatibility and automated quality control (Yamagata et al., 4 Aug 2025).

In that form, the SSBD Ontology is best understood as a semantic framework that links organism → biosample → imaging method → image URI, while simultaneously preserving publication context, dimensional metadata, and cloud-native access paths. Its distinctive contribution lies in the combination of two-tier curation, REMBI alignment, OBO/FBbi interoperability, and an operational RDF/OWL deployment that supports cross-dataset semantic discovery in bioimaging (Yamagata et al., 4 Aug 2025).